Hopefully that benefits section was immediately followed by a caveats/downsides section!
floats have a meager 6 digits of precision
doubles have a better 13 digits of precision, but still far away from the 38 given by number type
floats and doubles are incapable of representing *many* numbers - they just cannot hold them.
ops$tkyte%ORA11GR2> set numformat 9999.9999999999999999999999999999999999999
ops$tkyte%ORA11GR2> select 1.0101f, 1.0101 from dual;
1.0101F
-------------------------------------------
1.0101
-------------------------------------------
1.0101000100000000000000000000000000000
1.0101000000000000000000000000000000000
In SQL, the benefits of native operations don't really show up unless you are using very specific functions - like LN(). for example select sum(float) vs select sum(number) will not be very different (but select sum(ln(float)) will outperform select sum(ln(number)) greatly)
In PLSQL, if you are doing lots of computations - you'll find that floats/doubles outperform numbers - sometimes by a lot. But remember - lack of precision, inability to store certain numbers altogether make them useless for financial applications and the like.
You have to be *very* careful using them.