Your updated numbers are okay, but your conclusion is crap.
Of course Caltech looks at race, but unlike Harvard, it's a more of a STEM pool that it's drawing from, so it looks like the STEM pool of applicants. In fact when I attended, often admissions reviewers (who were mostly white-male faculty) would openly lament the multitude of applications from the asian-with-glasses-plays-piano-or-violin stereotype.
FWIW, the big push over the last 20 years at Caltech was to increase the number of women attending (not different races), so it is totally unsurprising (to me) that the asian population went up since the time I attended because it probably just mirrors the female-STEM high school population in California.
As for graduating on time, well, it's a tough school, but statistics don't bear out your assertion that rich asians are under-represented in the didn't-graduate-in-4-years-or-at-all pool. When I attended there, one of my school newspaper reporters looked at the statistics for an in-depth admission report and there is very little correlation with anything in the didn't-graduate-in-4-years rate (other than parents being a California resident).
Most potentially negative educational outcomes often were somewhat correlated with being a California resident (even though being private school, there is no tuition break for a California resident), because not being a well-known national institution (but more even with Harvard internationally in the STEM area, but behind MIT) the student pool is somewhat self-selecting having more qualified folks be out-of-state and international students. Also most other things being equal, California residents matriculate at a higher rate (east coasters accepted to multiple schools matriculate at Caltech at a much lower rate), and more likely to transfer to a UC school before they graduate if they are having trouble (rather than stick-it-out).
Of course the fact that California-resident Techers are also more Asian than the out-of-state/international pool, creates some Asian correlation in every statistic. However, with an entering class size of only about 200-250 people, the statistical significance of any trend is probably dubious as well. According to federal statistics, in 2006, 100% of underrepresented minorities graduated, but this fell to 80% in 2007, I think that was a result of 1 person less.
Although 80% average graduation rate means 40 folks don't graduate on-time, but in my class, I know of 10 folks that actually flamed out after 1 year. Also, if you look at the statistics of the student residence houses, the highest graduation rates are Lloyd and Fleming house and the lowest graduation rates are Dabney and Avery. Knowing the typical populations of these houses, I would say that a generalization that somehow accepting more Asians would help bring up the graduation rate is somewhat dubious (Avery and Lloyd house might be a poster-child for asian high-achievers, Fleming is much more representative of the population at large, and Dabney probably isn't representative of any population that is tracked by admission statistics). If manipulating statistics was the goal, Caltech should be simply accepting fewer Californians (which it's been trying to do for quite a while)...