Plus, fractional computing has severe problems with expanding denominators, which grow very quickly with non-trivial operands. Although computers are fast, the overhead of dealing with hundred-digit-long arbitrary precision integers eventually makes rationals unwieldy.
I'd rather make that trade-off myself than have the language decide for me. Most of the time I would start with correctness and if speed is needed and the profiler shows that the bottleneck lies in fractional arithmetic, I'll switch to floating point numbers.
A similar trade-off exists with bignums versus 32 or 64 bit integers. There again, I prefer a language that uses bignums by default. I'll convert the bottlenecks to use fixnums after I've verified that it is safe to do so.