2.5 degrees is about 245 km. Without the boostback, the stage ends up coming down about twice that distance to the east of the launch site despite that motion. But that's missing the point, particularly for RTLS: the directions considered horizontal and vertical are basically identical at launch and landing, and adding to vertical velocity after staging will only send the stage further downrange (unless you posit spending utterly unrealistic amounts of delta-v on the maneuver). RTLS is not an orbital maneuver, and the reasons for those rules of thumb you quoted do not apply. And yes, of course SpaceX accounts for Earth's rotation and so on, they need to actually hit a precise landing target. However, they are irrelevant to understanding the basic maneuvers, and simply do not require the maneuvers you describe.
Another factor is that if things were done as you say, the RTLS would take far, far longer than the ASDS landings, because the stage would have to be thrown on a high enough trajectory that the landing site could come into position under it, which would take much more of a rotation. The fact that CRS-9 did the entry burn just 4 minutes after the boostback burn should be another hint that something's wrong with your analysis.
Since you apparently trust Flight Club, just look at their CRS-9 model: https://flightclub.io/results/...
The boostback burn happens from 162 to 211 s. For the first 7 seconds of that, it is completing a maneuver to a pitch angle of -2 degrees, where it remains for the remainder of the burn.