I had the exact problem on a 16F627A. It was caused by entering a routine at power-up
that wrote to EEPROM.

The cure was as simple as adding a short delay at power-up to allow things to stabilize.

See if adding a short delay at power-up doesn't cure this.