P.S.: @pdmtl guys
It's plain wrong to have a wet/dry parameter for dynamic processing fx.
It just doesn't make sense at all to have the compressor output mixed
with the input signal (It not only doesn't make sense, it even adds
strange phasing effects, if the the dynamic processor uses a look-ahead
delay).
Can we agree on that? And if not, can we discuss this, so that we
finally can agree on that?

That sounds like parallel (or New York) compression, which is far from being wrong. It allows for an increase in RMS without affecting the source's transient response, and in many cases this technique is far superior to series compression. A fair majority of rock/pop records of the past quarter century have had parallel compression applied to the drum bus.

In the box, latency has to be compensated for though, so you will have to delay the source to properly mix with the compressed output. You can simply send the 'dry' signal through the same compression with a 1:1 ratio and high threshold to achieve this.

http://en.wikipedia.org/wiki/Parallel_compression