If you render to a surface that is less than 1/2 of the GPUs tile size in either dimension, the driver has to allocate a strided surface for the render and then copy from the the strided buffer to the non-strided buffer at the end of each render. This shouldn’t have a significant impact on your performance. Depending on your bottlenecks though, you may see a performance gain by increasing the FBO attachment dimensions to match the GPUs requirements, e.g. 32×8 or 32×16 depending on the GPU you are targeting.
As a rule of thumb, we would recommend allocating a small number of large textures that you render many objects into rather than rendering to many small textures.