With special-casing for the first two strokes, we could probably use the following rules to continue bashing:
3a) There is removable terrain withing the next 14 pixels (and at the correct height)
3b) There is steel within the next (14-5*n) pixels (and at the correct height), where n is the number of previous basher stokes that haven't removed any terrain.
3c) There is steel within the next (19-5*n) pixels (and at the correct height), where n is the number of previous basher stokes that haven't removed any terrain, and where the basher has made at most 3 strokes.

There are still some problems with this:
- In situation i) above, one can manage about four basher strokes before removing terrain for the first time.
- I can still cook up situations where terrain behind a steel wall determines whether a basher turns or not. Most notably this occurs when putting a steel wall (with possibly some usual terrain behind it) after a steel slops (like in möbius' example).
Moreover I am very worried that such complex rules like the above might introduce new glitches, that only occur in very specific situations and on very specific basher strokes.

Unless someone has a brilliant new idea, I therefore tend to choosing some basher mechanics that fulfil only two of the following three conditions:
C1) Bashers turn at steel only when standing directly in front of it, cf. http://www.lemmingsforums.net/index.php?topic=2579.0
C2) Steel-bashing to turn climbers is possible and it does not depend on the exitance of normal terrain behind the steel wall, cf. first post in this thread.
C3) On steel slopes, bashers do not continue bashing, cf. the example level that möbius posted in this thread.

Presonally, I am in favor of keeping the current V1.47 behavior and accepting that C3) is not satisfied. Reason:
- Not-C1) does have too many weird consequences and whether a basher turns depends strongly on the pixel they started bashing.
- "Terrain behind steel"-influences are less discoverable by players than the C3)-behavior. In other words: Players can probably guess what conditions need to be satisfied for Not-C3) to happen, once they see it. Guessing in what setups the "terrain behind steel"-influence comes up (without already knowing the cause) is far harder. 
Therefore I deem ignoring C3) the most user-friendly option.
Note that the Lix-basher satisfied C2) and C3), but not C1).

Just uploaded a new version of GSTool. It supports some settings that don't work in the current version but will in future versions, most notably, a new object type: "Background Image". An object set to this (one per graphic set) will be used as a background image on any level that uses the graphic set in question as its primary graphic set.

Most other newly-supported features are for long-term purposes, not immediate future.

Note that colors work a bit differently with this new version. You no longer set particle colors; instead, you can set the mask color (used eg. for bridge pixels), minimap color, and background color; with the remaining colors being unused (but the first three have planned future use; GSTool, when opening an old DAT graphic set, will automatically set the colors correctly for the new uses, including these future ones. I'm not sure if it sets them correctly when opening an old graphic set from any other format.)

Updated the experimental again. Includes Nepster's recent physics fixes, a fix for the issue of new version's settings overriding old version's, a fix for the crash on levels with custom window orders, and custom sound effects now work.

Phew, I just solved Harkonnen Force. That was really hard, but it made me feel smart after solving it
I also beat Arty's levels which were both fun. I especially liked the Banjo-Kazooie themed one. I beat Wishing well, but my solution was very similar to Arty's latest back-route solution. Still a cool level. I really like the moss. Attached is the Harkonnen Force replay. I won't attach the rest of the replays because they're all very similar to ones already posted.

Spoiler (click to show/hide)

On Harkonnen Force, I manage to bypass completely the trap at the start.