I got to thinking about this thread recently when I ran a replay of the 1922 season playing the St. Louis Browns (won the pennant
lost the world series
). In the League Settings, I set the starting rotation size to 6 and number of relievers to 1. I set Starting Rotation Mode to "Start Highest Rested" and allowed starters in relief. My reasons for doing so are outlined in the posts above. Here are the results for the top five Brown starters, listed in their order in the rotation, comparing the sim totals to real life:
Urban Shocker (sim): 50 G, 26 GS, 274 IP
Urban Shocker (RL): 48 G, 38 GS, 348 IP
Elam Vangilder (sim): 38 G, 31 GS, 285 IP
Elam Vangilder (RL): 43 G, 30 GS, 245 IP
Ray Kolp (sim): 35 G, 24 GS, 222.1 IP
Ray Kolp (RL): 32 G, 18 GS, 169.2 IP
Dave Danforth (sim): 30 G, 20 GS, 166 IP
Dave Danforth (RL): 20 G, 10 GS, 79.2 IP
Dixie Davis (sim): 29 G, 28 GS, 227 IP
Dixie Davis (RL): 25 G, 25 GS, 174.1 IP
The fear, then, that this set-up would result in too many pitchers logging 300+ innings wasn't borne out. In fact, while three pitchers totaled over 300 innings in real life (Red Faber - 352; Urban Shocker - 348; Eppa Rixey - 313.1), no player in my replay reached that level (Jesse Haines led the majors with 292.2). Indeed, it could be argued that my set-up led to pitchers - at least the top starters - pitching too few innings.
The main reason, as I noted earlier, is that the game uses the top starter as the top reliever as well. So Shocker, my top starter, came out of the bullpen 24 times during the season. That meant that he missed starts because he was frequently tired, so starts that would have gone to him went instead to starters farther down the list. The biggest beneficiary of that was probably Ray Kolp, the third starter, who got six more starts in the sim than he did in real life (Danforth started the year in the bullpen but moved to the rotation because of injuries and stayed there because he proved to be better than the guy he replaced - in real life he was demoted to the minors mid-season because he was suspected of doctoring the baseball).
That was the case as well with the AI teams. Haines, the top starter for the Cardinals, started 30 games and relieved in 21. In real life, he started 26 games and relieved only 3 times. The second pitcher in the Cards' rotation, Bill Doak, started more games (32) than Haines (30). That pattern repeated itself on a number of teams, where the second or third starter in the rotation logged the most starts on the staff but the top starter logged the most innings pitched because of all of his relief appearances.
As I argued above, the best solution for the pre-reliever era (roughly pre-1940) would be scrap the existing rotation-based pitching system and replace it with something that would recognize that pitchers in this era were expected to start and relieve. Barring that (and I have no realistic hopes of that ever happening), the next-best solution would be to make the starting rotation equal the number of pitchers on a staff, so that a team with a seven-man staff would have a seven-man rotation. That's not possible right now because the rotation size is limited to 6, but it would presumably be easier to change the rotation limit than scrapping the whole pitching system and starting over from scratch.