Even on the current hardware, floats are very efficient.
Even if the objects communicates with fixed point q27 (frac32),
you can use floats inside the objects as long as you translate them
back and from the with instructions like
These are efficient.
Floats inside objects are very efficient because there is 32 float registers in the FPU. This means that local float variables won't be stored in SRAM but in registers.
Fixed point fractional are very useful when you deal with accessing and interpolating tables (addressing, modulo a power of two and use of 1 cycle mul acc).
In many occasions i optimized my objects with a mixture of floats and fractionals.