Go for something with a simpler instruction set like a 6502 or a 6809. The 8086 and z80 are practically siblings (read the history of the z80) and have a larger instruction set to implement.
I assume you're going to do a scratch registers + mutliplexor/demultiplexor + ALU with microcode solution?
(I'm no expert, the only uC I ever implemented myself [in hardware that is] was in uni using EPROMs and PALs, because I'm ancient ;-p )
Edit: I know that doesn't address your desire for floating point, but I can't think of a simple processor that has it. Was going to suggest Arm32 as well, but again, no floating point by default.
Maybe an 8086/8087 combo really is the best/simplest choice, but it's still a decently large instuction set, especially when you have to implement the system level instructions too.