According to the documentation from the ArcGIS site, Cost Allocation
Calculates for each cell its nearest source based on the least
accumulative cost over a cost surface.
You ended up with only 14 spaces because no cells have those four sources as their nearest source in terms of accumulated cost (as indicated by your cost raster). Had this been simple geometric proximity allocation, then you would have 18 spaces. However, nearness here is expressed in terms of accumulated travel cost. You learn more about how the computation is done here.
I think what you expected were Thiessen polygons.
To address @whuber's points, I think we need the original data sources ( input rasters, cost rasters ) to do a more specific analysis. Tried to replicate the analysis but as I only had the source points as opposed to the original raster, my results weren't the same. I got the following cost allocation raster: