Version:            (using KDE 4.3.1)
Installed from:    SuSE RPMs

Floating-point calculation always have some slight error caused by the precision of the binary representation of the number.
KRunner shows too much decimal digits, thus showing to the user these slight errors.
I think 6 decimal places or so are sufficient, allowing enough precision for most application and hiding small internal errors.
If you really want to keep this number of decimal places, you should use a higher-precision floating-point type, like «long double» instead of «double» or so.

Steps to reproduce:

Enter 190.12 - 186.69 =

Expected result:

= 3.43

Actual result:

= 3.430000000000007