[nat]

For recent seasons WAR uses play-by-play data, which gives you precise information about exactly where the ball was hit. Obviously that's not available for older seasons. For those seasons it uses data comparing plays made versus average numbers of plays made by other players at the same position. (Presumably with adjustments for the field the guy plays in - outfielders in big parks have a harder time, for example.) I don't have the details handy, but because the older data is less reliable than the current data it's also regressed to the mean to some degree. This means that really good and really bad defensive performances are going to be smoothed out a bit.

Given the limitations on the data, this is the best we can do. But it's not too bad. For both recent and old players their defensive contributions are on the same scale (we can estimate runs saved, and hence games won, from both sets of data). For older players you should take their defensive numbers to be an estimate with a bit of uncertainty on either side. As with anything statistical, how much uncertainty there is depends on how big of a sample you're talking about - for individual seasons its kind of iffy, for whole careers it'll be much better.

Clemente, by the way, was awesome. By Rfield even Ozzie Smith has only one season better than Clemente's best. Most of the guys on the career dWAR list ahead of him were middle infielders or catchers (because they get a big positional bonus). Curiously, he's exactly tied with another right fielder with a legendary arm, Jesse Barfield.

Third basemen get a small positional bonus (about 2 or 3 runs per season). You can find the adjustment on baseball-reference under the column labeled Rpos.