sos and rpi are horribly skewed in non conference play. part of it is small sample size, but that would average out over time. worse, there is a skew against the quality teams, and for the lowest quality teams - i.e. the sims (and thus, those who play them). think of it like this - if you had 3 pools of teams, each third by quality, and they only played each other, then when you calculated their rpi/sos, the range and distribution would be roughly equal for each third. so for example, on average, 3 of the top 9 teams by rpi would be in the bottom third of teams by quality.
well, in reality, the top 30 or so programs in a division play each other a huge % of the games in non conference. on the bottom end, sims, and teams who have not scheduled (who are usually among the worst), end up playing each other a disproportionately large amount of the time. so you basically get a similar effect to what i described in the first paragraph, albeit less extreme because there is some cross over between the groups.
the team you are talking about has played mostly sims with solid records. so, they are basically at the very top of the skew, and their 21 sos is practically meaningless (or, it will be by season end).
also, just FYI, another team having a strong SOS does not really help your rpi. your rpi is 25% your w/l, 50% your opponents w/l, and 25% your opponents opponents w/l. sos is the same thing, without the 25% that is your w/l. keep in mind w/l is weighted, 1.4 wins per road win, 1.4 losses per home loss, and .6 wins for a home win, .6 losses for a road loss.
anyway, your opponents sos is 2/3rd their opponents w/l, which is your opponents opponents w/l, which is only 25% of your rpi formula. the other 1/3rd of your opponents sos does not factor into your rpi at all. rather, you should be much more concerned with your opponents w/l, as that is half your rpi equation. which, is also strongly in your favor at this point. but again, that is partly because they played such a weak (in true terms) schedule so far - so their record can very easily drop off over the rest of the regular season.
disclaimer: i just realized this was d3. in some d3 worlds, where there are like 50 humans, and like 300+ sims, the same concepts apply, but not exactly like i described. i described it from the "normal" scenario standpoint (mostly humans, 1/3rd sims or so or less), or at least, what used to be normal =(