diff --git a/src/core/cpu_recompiler_code_generator_aarch32.cpp b/src/core/cpu_recompiler_code_generator_aarch32.cpp
index 2cb83f2ba..397a75263 100644
--- a/src/core/cpu_recompiler_code_generator_aarch32.cpp
+++ b/src/core/cpu_recompiler_code_generator_aarch32.cpp
@@ -1276,8 +1276,6 @@ void CodeGenerator::EmitLoadGuestMemoryFastmem(const CodeBlockInstruction& cbi,
       break;
   }
 
-  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(Bus::RAM_READ_TICKS));
-
   bpi.host_code_size = static_cast<u32>(
     static_cast<ptrdiff_t>(static_cast<u8*>(GetCurrentNearCodePointer()) - static_cast<u8*>(bpi.host_pc)));
 
@@ -1286,8 +1284,17 @@ void CodeGenerator::EmitLoadGuestMemoryFastmem(const CodeBlockInstruction& cbi,
   // generate slowmem fallback
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
+
+  // we add the ticks *after* the add here, since we counted incorrectly, then correct for it below
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+  m_delayed_cycles_add += Bus::RAM_READ_TICKS;
+
   EmitLoadGuestMemorySlowmem(cbi, address, size, result, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // restore fastmem base state for the next instruction
   if (old_store_fastmem_base)
     fastmem_base = GetFastmemStoreBase();
@@ -1436,8 +1443,15 @@ void CodeGenerator::EmitStoreGuestMemoryFastmem(const CodeBlockInstruction& cbi,
   // generate slowmem fallback
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
+
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+
   EmitStoreGuestMemorySlowmem(cbi, address, size, actual_value, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // restore fastmem base state for the next instruction
   if (old_load_fastmem_base)
     fastmem_base = GetFastmemLoadBase();
diff --git a/src/core/cpu_recompiler_code_generator_aarch64.cpp b/src/core/cpu_recompiler_code_generator_aarch64.cpp
index 52f3fc32e..178b629cf 100644
--- a/src/core/cpu_recompiler_code_generator_aarch64.cpp
+++ b/src/core/cpu_recompiler_code_generator_aarch64.cpp
@@ -1463,16 +1463,23 @@ void CodeGenerator::EmitLoadGuestMemoryFastmem(const CodeBlockInstruction& cbi,
     }
   }
 
-  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(Bus::RAM_READ_TICKS));
-
   bpi.host_code_size = static_cast<u32>(
     static_cast<ptrdiff_t>(static_cast<u8*>(GetCurrentNearCodePointer()) - static_cast<u8*>(bpi.host_pc)));
 
   // generate slowmem fallback
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
+
+  // we add the ticks *after* the add here, since we counted incorrectly, then correct for it below
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+  m_delayed_cycles_add += Bus::RAM_READ_TICKS;
+
   EmitLoadGuestMemorySlowmem(cbi, address, size, result, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // return to the block code
   EmitBranch(GetCurrentNearCodePointer(), false);
 
@@ -1638,8 +1645,14 @@ void CodeGenerator::EmitStoreGuestMemoryFastmem(const CodeBlockInstruction& cbi,
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
 
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+
   EmitStoreGuestMemorySlowmem(cbi, address, size, value_in_hr, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // return to the block code
   EmitBranch(GetCurrentNearCodePointer(), false);
 
diff --git a/src/core/cpu_recompiler_code_generator_generic.cpp b/src/core/cpu_recompiler_code_generator_generic.cpp
index 73b146b62..2585f1f29 100644
--- a/src/core/cpu_recompiler_code_generator_generic.cpp
+++ b/src/core/cpu_recompiler_code_generator_generic.cpp
@@ -57,8 +57,6 @@ Value CodeGenerator::EmitLoadGuestMemory(const CodeBlockInstruction& cbi, const
     }
   }
 
-  AddPendingCycles(true);
-
   Value result = m_register_cache.AllocateScratch(HostPointerSize);
 
   const bool use_fastmem =
@@ -83,6 +81,7 @@ Value CodeGenerator::EmitLoadGuestMemory(const CodeBlockInstruction& cbi, const
   }
   else
   {
+    AddPendingCycles(true);
     m_register_cache.FlushCallerSavedGuestRegisters(true, true);
     EmitLoadGuestMemorySlowmem(cbi, address, size, result, false);
   }
@@ -133,8 +132,6 @@ void CodeGenerator::EmitStoreGuestMemory(const CodeBlockInstruction& cbi, const
     }
   }
 
-  AddPendingCycles(true);
-
   const bool use_fastmem =
     (address_spec ? Bus::CanUseFastmemForAddress(*address_spec) : true) && !SpeculativeIsCacheIsolated();
   if (address_spec)
@@ -157,6 +154,7 @@ void CodeGenerator::EmitStoreGuestMemory(const CodeBlockInstruction& cbi, const
   }
   else
   {
+    AddPendingCycles(true);
     m_register_cache.FlushCallerSavedGuestRegisters(true, true);
     EmitStoreGuestMemorySlowmem(cbi, address, size, value, false);
   }
diff --git a/src/core/cpu_recompiler_code_generator_x64.cpp b/src/core/cpu_recompiler_code_generator_x64.cpp
index d08c766a6..44592435a 100644
--- a/src/core/cpu_recompiler_code_generator_x64.cpp
+++ b/src/core/cpu_recompiler_code_generator_x64.cpp
@@ -1955,9 +1955,6 @@ void CodeGenerator::EmitLoadGuestMemoryFastmem(const CodeBlockInstruction& cbi,
     }
   }
 
-  // TODO: BIOS reads...
-  EmitAddCPUStructField(offsetof(CPU::State, pending_ticks), Value::FromConstantU32(Bus::RAM_READ_TICKS));
-
   // insert nops, we need at least 5 bytes for a relative jump
   const u32 fastmem_size =
     static_cast<u32>(static_cast<u8*>(GetCurrentNearCodePointer()) - static_cast<u8*>(bpi.host_pc));
@@ -1972,8 +1969,17 @@ void CodeGenerator::EmitLoadGuestMemoryFastmem(const CodeBlockInstruction& cbi,
   m_far_emitter.align(16);
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
+
+  // we add the ticks *after* the add here, since we counted incorrectly, then correct for it below
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+  m_delayed_cycles_add += Bus::RAM_READ_TICKS;
+
   EmitLoadGuestMemorySlowmem(cbi, address, size, result, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // return to the block code
   m_emit->jmp(GetCurrentNearCodePointer());
 
@@ -2234,8 +2240,14 @@ void CodeGenerator::EmitStoreGuestMemoryFastmem(const CodeBlockInstruction& cbi,
   bpi.host_slowmem_pc = GetCurrentFarCodePointer();
   SwitchToFarCode();
 
+  DebugAssert(m_delayed_cycles_add > 0);
+  EmitAddCPUStructField(offsetof(State, pending_ticks), Value::FromConstantU32(static_cast<u32>(m_delayed_cycles_add)));
+
   EmitStoreGuestMemorySlowmem(cbi, address, size, value, true);
 
+  EmitAddCPUStructField(offsetof(State, pending_ticks),
+                        Value::FromConstantU32(static_cast<u32>(-m_delayed_cycles_add)));
+
   // return to the block code
   m_emit->jmp(GetCurrentNearCodePointer());