If you had a different height and width then the totals over all of the glcm results would be different. Whether that would be a problem would depend upon what you did with the information. For example if you normalized it by dividing the counts by the totals then once the image was large "enough" to get some accuracy the exact numbers might not matter.
The original JPEG did not have a lossless compression format, but those are often implemented as an extension. When you talk about sizes like that you are hinting at different quality levels of JPEG compression, not about lossless JPEG.
One thing you need to know about JPEG is that it is weak on sharp edges, and the lower your quality (the smaller the file) the weaker it gets. For pictures of (for example) clouds that doesn't matter much, but if you are trying to do texture analysis then it could be crucial. If your GLCM is being done at a distance of 1 (adjacent pixels) then you could have substantial differences due to JPEG approximating edges.
