If you overlay an EIT 304 image and an AIA 304 image the circumferences of each disk are almost cospatial. This indicates that the pointing in both FITS files and the overlay algorithm are working well. If you do the same thing in hv.org, you find that both images appear to be centered well, but the EIT sun appears to be larger than the AIA sun. The contrasting results suggest that the problem lies in how each application calculates the image to display. JHV relies on each instrument's FITS files containing some measurement of the solar radius as seen by that instrument. Since the solar radius is a known and fixed number, JHV uses that to scale the image appropriately. Since all images are scaled to the same fixed solar radius, the disks from all imagers appear to be the same size.
HV.org does not do this; instead, it uses the values of CDELT1 and CDELT2 along with other positioning information in the FITS header in order to calculate which sections of the data to show, and where to put them. The problem with doing this is that EIT fixes the size of the CDELT1 and CDELT2, but changes the measured radius of the Sun as stored in the FITS files appropriately (since the Sun-SOHO distance varies with time, the apparent radius of the Sun changes in the image). A suggested solution is for hv.org to implement the scaling algorithm of JHV in order to better co-align images in hv.org. This may also help the co-alignment of LASCO C2 and C3 data, which also appears to be better in JHV than in hv.org. Thanks!