This is a bug in the internal sun.security.krb5.internal.ktab package.
KrbTabInputStream#readEntry(length,kvno) reads length of bytes into a KeyTabEntry. A variable "index" is used to store how many bytes are left unread to determine if there are enough (or more) for this entry. The variable should be per-read or at least per-instance (since a KrbTabInputStream should be read sequentially and thus not multi-thread enabled). However, it's per-class (static) now.
This means when multiple threads are calling the method at the same time (for example, both refreshing a keytab), the value might be modified by another thread during the execution of the method. If it goes bigger, a false skip() will be called, otherwise, it triggers a keytab corruption error. Both lead to the keytab singleton object to be null.