I was asked the other day about how to load multiple x file -based models into VE3D without having to place each one manually. I've written up a sample here. The idea is more or less a fusion of the existing ActorDataSource and XFile samples, using the actor from the XFile sample rather than the bunnies. I've also mocked up a very simple data file that contains placement information, rather than generating data on the fly as for the bunnies.
The way I've done this, the parsing code for the x files is done in the actor. In this way, swapping out file formats should be straightforward so long as you have parsing code for it (sorry, only x files are directly implemented in the engine, but the results of any parse should still be easily convertible into VE3D graphics objects).
For the sake of simplicitly I did do one thing wrong. The flow of the code is like this: multiple background threads all execute QueryPrimitivesInternal at the same time, and when they are finished, they move through a lock into ActorBuilder. They are still on background threads, but they now execute serially. So, really the data retrieval (in this case, loading from resources, but probably a network request) and the parsing, as appropriate, should happen in QueryPrimitivesInternal, and the action in ActorBuilder should be kept to a minimum. I violated this rule in the sample code for the sake of illustration, but you should be aware of it.
Finally, an important consideration is level of detail. Dense and numerous models can slow down loading and rendering. DataSources take advantage of the ActorBounds and scale concepts, such that it is possible to use simplified models at certain LODs, and switch to more detailed as the user zooms in. If your needs are less extreme, then you can simply use scale to cause models to stop rendering when not in view, and at a certain distance, which is what I have done in the sample.