This system was invented in 1973 by Clifford Christopher Cocks of the GCHQ and described in his document (then secret) A note on Non-Secret Encryption, before RSA was published. See this question for more context.
Is this scheme more secure or less secure than the normal RSA? Better or worse?
On security, we know no difference with RSA. The restriction that $p$ and $q$ are relatively prime to $p-1$ and $q-1$ does not sizably reduce the keyspace, in fact this restriction is always met when $p/2<q<2p$, which is customary in RSA as practiced. And, by using $N$ as the public exponent, nothing more than their product is revealed about the factors. From this standpoint, RSA with small exponent reveals more: with $e=3$, about half the primes are ruled out, and this is not known to make factoring sizably easier. But, like in RSA, factorization is only one of the security threats: we know no reduction to factorization for either scheme.
On practicality, RSA is superior because it can use a small $e$, which greatly speeds-up encryption (and signature verification), with no known drawback if proper padding and leakage protection is used, or $e$ is comfortably above the bit size of $N$. Small $e$ was not in Ronald L. Rivest, Adi Shamir, and Leonard Adleman's A Method for Obtaining Digital Signatures and Public-Key Cryptosystems, in Communications of the ACM, Feb. 1978. But that quickly was added by Rivest's MIT group: it's in the challenge they made for Martin Gardner's A new kind of cipher that would take millions of years to break (in the Mathematical Games column of Scientific American, Aug. 1977).
Cock's paper clearly suggests using the Chinese Remainder Theorem for decryption, which from the standpoint of speed is superior to RSA as initially described. As far as I know, that improvement was not published before Jean-Jacques Quisquater and Chantal Couvreur's Fast decipherement algorithm for RSA public-key cyptosystem, in Electronics Letters, Oct. 1982. It is now standard practice in RSA, because it speeds up things by a factor up to about 4 compared to computing $x^d\bmod N$ by performing all the arithmetic modulo $N$.