The problem is not the "diagonal bullet", but "the bullet". At that moment, there is some hidden objects that loads the CPU (the owls inside the stones). The bullet is the object that break the frame limit and then, the game runs at 30fps.
Umh, however, maybe there is not too much road in gettng a 3.5Mhz version.
Actually it run well at 5.x mhz. To have an idea of how much you need to optimize, maybe you can try it on openMSX with 4Mhz for example and check for slowdowns. if it is slow try 4.5mhz and so on.
When you see that there is no slowdown anymore, you can get an idea of how much you need to optimize. For example say that @4.1Mhz there is no slowdown. You need to reduce the n. of cycles of 3.5/4.1 for each frame. Based on this value you can optimize better some or another parts, or reduce the number of moving objects when on 3.5Mhz.
You are not too far from the goal the game run on 5Mhz already.
