In some cases it's better to use a single larger image for your sprites.
It will more likely result in less overhead if this image is to be cached
Some hw configurations only allow power of two textures, so
for example your 24x35 sprite will have to be placed into a 32x64 texture.
So if you have tons of textures like that your overhead will be
significant. But if you place all of them into a 512x512 texture,
it could be reduced.
Also, having your sptires in a single image may help with performance
as it requires less context switching for the native pipeline.
copy image 1 into texture 1
copy image 2 into texture 2
set texture 1
render texture 1 at coord1
set texture 2
render texture 2 at coord2
copy image into texture
render texture at coord1
render texture at coord2
And since setting a new texture is an expensive context switching
operaion (at least, for Direct3D), the second way is typically faster.
For software-only rendering it doesn't make that much
of a difference.