I'm using both, each for its own purpose.
I believe that most simple filtering queries you can make on PostGis have an equivalent on MongoDB (within, near, intersection). MongoDB has the additional advantage of being capable to ingest a sparse schema and nested properties (sure, pg can do this too, but you need to k ow beforehand which properties you will store as json). If you want to store the results of an api call to twitter or google places, you just need to identify the id and location fields and save the results as a document. No need to cast the geometry as it happens in PostGIS because it usually comes as geojson already.
However, if you're planning on performing any operation on geometries such as finding centroids, doing douglas peucker simplification, union, etc, you can't do that in mongo. Working with any srid different from 4326 (WGS84) you can't do in mongo. Aggregation is doable but not trivial.
Finally, postGIS offloads the burden of many operations to Geos and Gdal, which are proven tools that specialize in their tasks. MongoDB in turn performs its geo queries using its own logic. I'm sure mongo has amazing engineers, but it's a bit like reinventing the wheel.