The ECHECK 1000000000 value is insane; anything over 500 is usually a bad idea. Large energy changes typically indicate a problem with the model building that needs to be fixed, not suppressed.
DO NOT use TIMESTEP greater than 0.002 for atomistic simulations; our group uses and recommends 0.001 for most work. You should only increase NSTEP to get a longer simulation time for a given run, or use multiple runs with DYNA RESTART. See dynamc.doc
The above scripts indicate a method for which can be used to run much longer simulations using the RESTART capability of CHARMM, and some shell scripts to partially automate file numbering, failure checks, and continuation.
