A few years ago I assembled a similar database. The next step will be to compare my games with those from Chess960 Jungle and see if I have anything to contribute to the newer collection. There is also the important subject of whether or not such a database is useful for more than its historical overview.
I first collected games in August 2009, but don't seem to have blogged about it. All of those games were from the annual Mainz events, which I blogged about in Chess960 @ Chess Classic Mainz (2001-2008), and CCM9: Nakamura, Grischuk, and Rybka (2009).
Comparing my database with HarryO's, I discovered he was missing one event -- the 2004 Svidler - Aronian match -- and promptly forwarded it to him. The other events in our collections matched very well. I noted a few differences where he chose to exclude certain games. His first rule, 'Human ELO 2000+', excludes engine competitions, of which there were five at Mainz:-
2005; CCM5 - 1. Chess960 Computer WCh; 63 games
2006; CCM6 - 2. Chess960 Computer WCh; 90
2007; CCM7 - Livingston Chess960 Computer WCh; 21
2008; CCM8 - Livingston Chess960 Computer WCh; 34
2009; CCM9 - Livingston Chess960 Computer WCh; 34
His last rule, 'Simultaneous exhibitions where the games last at least 30 moves', limits the number of games from two early simuls:-
2003; CCM3 Chess960 Simul - Leko; 20
2003; CCM3 Chess960 Simul - Svidler; 20
There was another Mainz simul that he might have overlooked, perhaps because it was the only chess960 event organized in the final year of the Mainz series:-
2010; Simul - Kosteniuk; 20
The Mainz events are certainly missed, but there's no need to lament about the past. HarryO is right to include LSS correspondence games, and he might want to consider ICCF games, where the pool of active players is probably stronger than for LSS. I mentioned the site's chess960 activity once in Correspondence Chess Ratings and Chess960. As for other correspondence sites -- Chess.com is a popular example -- I don't know if the percentage of strong players is high enough to merit their inclusion.
Finally, we come to the question of how useful such a database is. With a little over a thousand games in the collection, it can't serve for any sort of opening preparation into one of the 959 non-traditional starting setups. The CCRL, with 100-150 examples per start position (SP) is also not particularly useful, because the positional thinking required in the opening is not the engines' strong point.
For a long time to come, the value of a chess960 database is going to be its de-facto record of chess960 competitions. This alone is enough to justify the effort involved in maintaining it. Imagine how much poorer chess history would be if the traditional chess games played prior to the great 1851 London tournament had never been recorded and saved.