Sorry for a longish (previous) post -- looks like what I've posted is sort of close to what current draft data model is anyway.
A few comments after reviewing the latter:
I'd agree that more use needs to be put onto acs_rels to handle project resource grouping. As mentioned above, I think that a good way to do it is to have resources pool assigned to a project, yet each task should get a separate resource assignment (including multiple resources to a single task, although this may mean that a task is not granular enough).
Dependancies: just an other-task-wise dependancy is not enough -- one may have tasks that are not dependant on any other tasks, yet do have constraints (like must-finish-before, etc.). In that sense `constraints' maybe a better word indeed.