1) when I created the BH smash, I didn't want to make it too weak, coz when you have to smash, you have very little time to do so, and you have to use some of this time to aim, thus I didn't want to penalize the players too much (the CPU could do both though)
2) Ok, I tried with 65% and it was better...
3) Lower Speedconversion = slower to change the run to the opposite direction = harder
If you're around here next year, try to catch the Beta of TEM2. As one of its main goals will be to get most possible realistic CPU vs CPU matches, your feedback will be useful !
While I know this thread is pretty old, I've spent at least the last five years tweaking various things in the tennis.ini to try to create a really solid emulation of real life tennis stats in CPU vs CPU matches. The biggest issue I continue to have the relationship between serving advantage, aces, and percentage of return points won.
To create reference ranges for things, I've researched on the World ATP site, which has archives of match stats included ace totals and return percentages. http://www.atpworldtour.com/
Additionally, there's a match statistics charting project you may or may not know about that breaks down every point in a match and keeps meticulous records on things like rally length, serve advantage, serve locations, winners, unforced errors, etc. http://www.tennisabstract.com/charting/
Based on what I've found on those two sites, generally, aces for a three-set match average around 2 or 3 to as high as 18 or 19 for bigger servers. Of course surface and serving play a role, but usually, the ace count stays in the single digits. Despite the lower ace count, return points won stays below 40% most of the time, even on clay (unless the return skills of both players is high). Rally length averages out at 4.2 shots per rally on the tour.
I have not been able to find that combination of stats through the game engine, either too many aces are produced (15-30 in a three-set match, on average) or the return percentage is too high (45 -50%). Rally length tends to stabilize with mtrun reducing speed by about 15-20%.
One suggestion that would go a long way in my estimation is a separate malus for the return of serve that is higher than the groundstroke rallies. That would emulate how a player must stretch to reach a well-placed serve, and create either a weak ball that is a winner on the server's first shot, or a forced error. That way the serving percentages are correct, but the ace count remains lower.