As a different solution (not necessarily better), we do the DITA to HTML transform in the browser. (This is also what Michael Kay does with DocBook, in his own way for his own docs.) We support a subset of DITA that is roughly equivalent to LwDITA, but it's all a matter of what you support in the XSLT. The one thorny issue is that we have to do a one-step transform... No pre/post passes. This has proved to be a problem with conrefs to things like table rows.
The good news is that it's very light-weight. On the server side, it's essentially a static site. It also seems to be quicker than calling the OT. In my ignorant innocence, I tried transforming single topics via the OT before turning to client-side transforms. Because of the multiple passes, the performance seemed sluggish. It could just be that I don't know the optimal way to invoke the OT.
Other good news is that we can integrate into our HTML GUI. So we use the same code to present microcontent in the GUI.
See it here: