There are only so many bits in a float. There are not always enough to represent the complete decimal expansion of certain numbers. As such, some decimal numbers are approximated when stored in memory. Doing arithmetic with an approximated float can sometimes cause rounding errors, which is what you are seeing.

Mandatory link in such thread: What Every Computer Scientist Should Know About Floating-Point Arithmetic
(One day, I will read it fully...)

Short version, as said above: floating-point calculations are inherently imprecise in the computing world, because of the way floating point numbers are stored. There is a necessary rounding.
You can use the double type if you really need more precision. Or BigDecimal if you don't care for speed and really need high precision.

Thanks, tfguy44. Although that does not quite clarify for me why 10 is a problem in that example whereas 1, 100, 1000, etc. are not (.1 etc. return 5.0E-4 etc.), it gave me the idea that I can work around the problem via nf().