The way to avoid these dramatic volume jumps is to use the Expression controller in conjunction with Velocity X-Fade. If you only use velocity X-fade, there will be an unavoidable jump in volume between the p and f layer. However, if you lower the Expression (which is really just a granular volume control) to "smooth" out the volume jump, you can - in almost every case - get the desired crescendo/diminuendo.
I like to think of velocity X-fade as the "rough" volume control and Expression as the "fine" volume control. Alternatively, you could think of velocity X-fade as "intensity" and Expression as "volume".
- Sam