diff --git a/src/engine/dispatch.h b/src/engine/dispatch.h
index fa57dda44..52e5c1c10 100644
--- a/src/engine/dispatch.h
+++ b/src/engine/dispatch.h
@@ -383,6 +383,8 @@ struct DivRegWrite {
    *   - xx is the instance ID
    *   - data is the sample position
    * - 0xffffffff: reset
+   * - 0xfffffffe: add delay
+   *   - data is the delay
    */
   unsigned int addr;
   unsigned int val;
diff --git a/src/engine/platform/fmsharedbase.h b/src/engine/platform/fmsharedbase.h
index ca5b52775..f8e0268dd 100644
--- a/src/engine/platform/fmsharedbase.h
+++ b/src/engine/platform/fmsharedbase.h
@@ -76,11 +76,11 @@ class DivPlatformFMBase: public DivDispatch {
     };
 
     struct QueuedWrite {
-      unsigned short addr;
+      unsigned int addr;
       unsigned char val;
       bool addrOrVal;
       QueuedWrite(): addr(0), val(0), addrOrVal(false) {}
-      QueuedWrite(unsigned short a, unsigned char v): addr(a), val(v), addrOrVal(false) {}
+      QueuedWrite(unsigned int a, unsigned char v): addr(a), val(v), addrOrVal(false) {}
     };
     FixedQueue<QueuedWrite,2048> writes;
 
@@ -97,7 +97,7 @@ class DivPlatformFMBase: public DivDispatch {
         pendingWrites[a]=v;
       }
     }
-    inline void immWrite(unsigned short a, unsigned char v) {
+    inline void immWrite(unsigned int a, unsigned char v) {
       if (!skipRegisterWrites) {
         writes.push_back(QueuedWrite(a,v));
         if (dumpWrites) {
diff --git a/src/engine/platform/genesis.cpp b/src/engine/platform/genesis.cpp
index 903d97a60..a2cfc7fbb 100644
--- a/src/engine/platform/genesis.cpp
+++ b/src/engine/platform/genesis.cpp
@@ -142,11 +142,16 @@ void DivPlatformGenesis::acquire_nuked(short** buf, size_t len) {
   for (size_t h=0; h<len; h++) {
     processDAC(rate);
 
+    if (delay>0) delay--;
+
     os[0]=0; os[1]=0;
     for (int i=0; i<6; i++) {
-      if (!writes.empty()) {
+      if (delay<=0 && !writes.empty()) {
         QueuedWrite& w=writes.front();
-        if (w.addrOrVal) {
+        if (w.addr==0xfffffffe) {
+          delay=w.val*3;
+          writes.pop_front();
+        } else if (w.addrOrVal) {
           //logV("%.3x=%.2x",w.addr,w.val);
           OPN2_Write(&fm,0x1+((w.addr>>8)<<1),w.val);
           regPool[w.addr&0x1ff]=w.val;
@@ -223,13 +228,19 @@ void DivPlatformGenesis::acquire_ymfm(short** buf, size_t len) {
 
   for (size_t h=0; h<len; h++) {
     processDAC(rate);
+
+    if (delay>0) delay--;
   
     os[0]=0; os[1]=0;
-    if (!writes.empty()) {
+    if (delay<=0 && !writes.empty()) {
       QueuedWrite& w=writes.front();
-      fm_ymfm->write(0x0+((w.addr>>8)<<1),w.addr);
-      fm_ymfm->write(0x1+((w.addr>>8)<<1),w.val);
-      regPool[w.addr&0x1ff]=w.val;
+      if (w.addr==0xfffffffe) {
+        delay=w.val;
+      } else {
+        fm_ymfm->write(0x0+((w.addr>>8)<<1),w.addr);
+        fm_ymfm->write(0x1+((w.addr>>8)<<1),w.val);
+        regPool[w.addr&0x1ff]=w.val;
+      }
       writes.pop_front();
 
       if (dacWrite>=0) {
@@ -372,9 +383,14 @@ void DivPlatformGenesis::acquire_nuked276(short** buf, size_t len) {
 
     //lleCycle=0;
 
-    if (!writes.empty()) {
+    if (delay>0) delay--;
+
+    if (delay<=0 && !writes.empty()) {
       QueuedWrite& w=writes.front();
-      if (w.addrOrVal) {
+      if (w.addr==0xfffffffe) {
+        delay=w.val;
+        writes.pop_front();
+      } else if (w.addrOrVal) {
         //logV("%.3x=%.2x",w.addr,w.val);
         //OPN2_Write(&fm,0x1+((w.addr>>8)<<1),w.val);
         was_reg_write=true;
@@ -871,9 +887,7 @@ void DivPlatformGenesis::tick(bool sysTick) {
 
   // hard reset handling
   if (mustHardReset) {
-    for (unsigned int i=hardResetElapsed; i<hardResetCycles; i++) {
-      immWrite(0xf0,i&0xff);
-    }
+    immWrite(0xfffffffe,hardResetCycles-hardResetElapsed);
     for (int i=0; i<csmChan; i++) {
       if (i==2 && extMode) continue;
       if ((chan[i].keyOn || chan[i].opMaskChanged) && chan[i].hardReset) {
@@ -1712,6 +1726,7 @@ void DivPlatformGenesis::reset() {
   dacWrite=-1;
   canWriteDAC=true;
   interruptSim=0;
+  delay=0;
 
   if (softPCM) {
     chan[5].dacMode=true;
diff --git a/src/engine/vgmOps.cpp b/src/engine/vgmOps.cpp
index 932155510..ad3599210 100644
--- a/src/engine/vgmOps.cpp
+++ b/src/engine/vgmOps.cpp
@@ -2686,9 +2686,15 @@ SafeWriter* DivEngine::saveVGM(bool* sysToExport, bool loop, int version, bool p
     // get register dumps and put them into delayed writes
     int writeNum=0;
     for (int i=0; i<song.systemLen; i++) {
+      int curDelay=0;
       std::vector<DivRegWrite>& writes=disCont[i].dispatch->getRegisterWrites();
       for (DivRegWrite& j: writes) {
-        sortedWrites.push_back(std::pair<int,DivDelayedWrite>(i,DivDelayedWrite(0,writeNum++,j.addr,j.val)));
+        if (j.addr==0xfffffffe) { // delay
+          curDelay+=(double)j.val*(44100.0/(double)disCont[i].dispatch->rate);
+          if (curDelay>totalWait) curDelay=totalWait-1;
+        } else {
+          sortedWrites.push_back(std::pair<int,DivDelayedWrite>(i,DivDelayedWrite(curDelay,writeNum++,j.addr,j.val)));
+        }
       }
       writes.clear();
     }