They should contain
{recordlength(=24)} {zr} {omega0} {flam0} {rho0b} {h} {boxL0} {recordlength=24}
{recordlength(=2097152)} 128*128*128=2097152 bytes of data ... {recordlength=2097152}
Most do have that form. But quite a few -- including most of the "rhob" files for times 101, 201, ..., 651, and all of the files for timestep 403, and many others -- have part of their first block zeroed out.
In fact, of the 3192 files of this type (399 timesteps * 8 data fields), fully 621 of them were corrupt in this way.
Here's a dump of the first part of an intact file:
[slevy@co-login1 MOVIEDATA]$ od -Ad -td4 movie3drhod.0797 | head 0000000 24 -1175056384 1050589266 1060152279 0000016 1027903249 1060152279 1106247680 24 0000032 2097152 -2139062144 -2139062144 -2139062144 0000048 -2139062144 -1182760832 -2138996556 -1363632212 0000064 -1699512174 -2134196300 -2086361432 -1264091008 0000080 -2137346892 -1266646860 -739843968 -2132691072 0000096 -875318144 -2137351516 -1010456155 -846482304 0000112 -908937790 -2137407555 -2139062144 -2139062144 0000128 -2139062144 -2139062144 -2139062144 -2139062144 0000144 -708607872 -2139062144 -2004844416 -2139062144
For comparison, here's a sample of a corrupt file. Note that the first record, with zr, ..., boxL0 are intact. But the remainder of the first 4096-byte block, including the "2097152" Fortran record header, are all zeros. Data from byte 4096 onward do seem to exist though.
[slevy@co-login1 MOVIEDATA]$ od -Ad -td4 movie3drhod.0403 | head 0000000 24 1069829152 1050589266 1060152279 0000016 1027903249 1060152279 1106247680 24 0000032 0 0 0 0 * 0004096 -252119826 -236462369 -619834358 319561984 0004112 -334755825 -1845955622 -2139062144 -2139062144 0004128 -2139062144 -2139062144 -2139062144 -843022208 0004144 -1311772712 -2066685770 -998189357 -1045251647 0004160 -1280606054 -2139058554 -1277837395 -909652864 0004176 -556609574 -1128086322 -689899045 -858136881
For some successes, see some histogram plots and renderings of 3-D field values.