I don't see how it is realistic to merge two OS source trees that have diverged over so many years. It was a very difficult and long drawn out process trying to merge some of the mess caused by the ANC and ART branches of RISC OS diverging over a much shorter timeframe.
I agree that open source is by no means a panacea - there are still many ways of killing it all off: e.g. lack of quality control, choosing an inappropriate licence such as GPL (unless it's dualled with modified BSD or similar), the remaining developer(s? getting bored and just going away.
I would say that the idea that other developers of other OSes would pinch ideas from the source is loopy, to be honest. RISC OS *in itself* has nothing to offer that other OSes don't already have in a far more sophisticated form already (for many years). The quoted example, the Window Manager, is a huge amount of ARM assembler that achieves relatively little in terms of functionality. It's a half-way house between a rectangle manager and a GUI environment and it does neither particularly well. It's major plus point for users is that it enforces various common traits in applications (icon bar positioning, window furniture behaviour), whilst that is also a weakness (lack of flexibility (cannot put the icon bar down the side of the screen instead of the bottom, for example), cannot customise behaviour of furniture easily, cannot use the furniture within your application). Thus you can get inter-operable applications with a common look-and-feel and that helps users, but ...
Users don't care about the inner workings of the OS directly - they *do* care about it indirectly, even if they don't know that they do. Users cannot and should not be expected to understand the inner workings of the OS. Consequently, I don't believe they realise just how awkward it is to write Wimp applications (particularly those that interact with the hardware) given the architecture of the OS.