Blowfish splits a 32-bit word into 4 bytes and insert each byte as an entry in a S-box.
No, it doesn't - Blowfish uses Sboxes with 32 bit entries. A specific Blowfish implementation may decide to insert each byte in a 4-byte Sbox entry separately, however it is not mandated to do so (and there are implementations that will insert all four bytes at once).
Does it consume the same cycles per byte as inserting a single byte?
That sort of question is hardware specific. On the other hand, for most anything above a microcontroller, writing four bytes in a single write of an aligned 32 bit word will be at least as fast (and on some platforms, faster) than an 8 bit write (and hence consume considerably fewer cycles per byte).
BTW: why do you care? Key scheduling (which in when generating sboxes are done, at least, for those ciphers with dynamic sboxes) is typically done once (when you get the key), and is usually not considered a performance bottleneck (at least, as ciphers are typically used).