Generally there is a slight increase if not more when anything is coded in C++. In the perlin noise plugin I've converted some of the intense and mostly used functions directly into assembler code, ie Noise and Interpolate. These two now actually use the FPU for all it's math instead of a built in function, even down to Cosine and the number 1.0. The DBP version now runs just as fast as what it would do in just C++ with these changes.
I did try converting the perlin noise to DBPro sometime ago (about two months), but whatever wasn't working at the time I couldn't find it and eventually went back to my C++ code.
As far as I know the plugin itself is compiled along with your program if it uses it. This plugin as it stands (er, give me a sec), 12.5k (Wow, I will have to reduce that!). That will get attached to your exe without the need to copy it.
Most of my programming is in C++ anyway but with the amount of help I've had from the community to date, my code will be released anyway.
Tonight I am considering changing the internal maths to single precision floats to give it another speed boost. Depends on my brain cell though.
I'm also aware of other noise algorithms which I'm thinking of implementing into the plugin or a separate one, such as the diamond-square, etc.
All in all, a plugin written directly in C++ will improve on speed of execution, mine has even been updated by the use of conversion of some functions into assembler.
Mental arithmetic? Me? (That's for computers) I can't subtract a fart from a plate of beans!
Warning! May contain Nuts!