Future high-resolution imaging X-ray observatories may require detectors with both fine spatial resolution and high quantum efficiency at relatively high X-ray energies (E ≥ 5 keV ). A silicon imaging detector meeting these requirements will have a ratio of detector thickness to pixel size of six or more, roughly twice that of legacy imaging sensors. The larger aspect ratio of such a sensor's detection volume implies greater diffusion of X-ray-produced charge packets. We investigate consequences of this fact for sensor performance, reporting charge diffusion measurements in a fully-depleted back-illuminated CCD with a thickness of 50 µm and pixel size of 8 µm. We are able to measure the size distributions of charge packets produced by 5.9 keV and 1.25 keV X-rays in this device. We find that individual charge packets exhibit a gaussian spatial distribution, and determine the frequency distribution of event widths for a range of detector bias (and thus internal electric field strength) levels. At the largest bias, we find a standard deviation for the largest charge packets (produced by X-ray interactions closest to the entrance surface of the device) of 3.9 µm. We show that the shape of the event width distribution provides a clear indicator of full depletion, and use a previously developed technique to infer the relationship between event width and interaction depth. We compare measured width distributions to simulations. Although we can obtain good agreement for a given detector bias, with our current simulation we are unable to fit the data for the full range of of bias levels with a single set of simulation parameters. We compare traditional, 'sum-above-threshold' algorithms for individual event amplitude determination to gaussian fitting of individual events and find that better spectroscopic performance is obtained with the former for 5.9 keV events, while the two methods provide comparable results at 1.25 keV. The reasons for this difference are discussed. We point out the importance of read-noise driven charge detection thresholds in degrading spectral resolution , and note that the derived read noise requirements for mission concepts such as AXIS and Lynx are probably too lax to assure that spectral resolution requirements can be met. While the measurements reported here were made with a CCD, we note that they have implications for the performance of high aspect-ratio silicon active pixel sensors as well.