The whole question comes down to how fast has it got to be. As you probably know, float operations take large amounts of code and data memory as well as many cycles to perform. I think you might need something with large program memory and a bit of ram. If you don't need high throughoutput this can be almost any mcu with sufficient memory. I've seen a PID temperature regulator operating on floats made with ATMega16, it could do some tens of iterations per second clocked at 16 MHz.
So if you are not after speed, when I'd say go with AVR. PIC18 could also do but has inferior processing power compared to AVR and the C18 compiler doesn't handle floats very well.
If you are after speed, then you could use something from PIC24H or dsPIC33 families. They operate at up to 40 MIPS which is a lot of processing power. The better choice would be dsPIC33 at it has some DSP functionalities which might come handy. I think ARM Cortex M-3 processors are also an option, as they have high processing power and usually lots of memory.
If you needed REALLY HIGH throughoutput you might need something with hardware floating point operation unit, but those might proove to be very expensive and hard in application. I can't think of anything specific right now (but that's probably because it's 6.50 in the morning right now... :/) but i guess TI and Renesas might have something like that.