I had never heard of this idea before. Do you have any references to this technique? Also, depending on what equations you're using, aren't you constrained to using a consistent system of units?
And yes you have to be consistent and stay in that unit system. The other remaining two we used plain Kelvin for temperature and for the charge Coulomb was measures in electron charge. That's it.
That shouldn't matter for the factors you're describing. The number of floats between 1 and 2 is the same as the number of floats between 2^-30 and 2^-31. Until you hit denormal numbers and start overflowing or underflowing your exponent, it doesn't have any effect on precision.
If you have a process that converts floats into other formats with more restricted exponents, like human readable strings, it might matter.
https://books.google.de/books?id=uzJbyD6_3DAC&pg=PA4
And yes you have to be consistent and stay in that unit system. The other remaining two we used plain Kelvin for temperature and for the charge Coulomb was measures in electron charge. That's it.