The principle is good and used in RSA-KEM. But the devil is in the details, specifically the conditions stated for security in the question:
the entropy bit string is of the same magnitude as the RSA modulus, or some number such that $m^e\gg n$
Problem is, we do not have a security reduction to the RSA problem from that condition. In RSA-KEM, $m$ is fully random over $[0,n)$, and the symmetric key is derived from $m$. In this way, we have security directly reducible to the RSA problem.
This issue of having no proof is not only theoretical: there would be serious danger if we used the proposed method to transfer a 128-bit uniformly random $m$ encrypted as $c=m^e\bmod n$ by an RSA public key $(n,e)$ with $e=65537$, which matches the stated criteria. There is a theoretical Meet-in-the-Middle attack with cost about $u+v$ modular operations which recovers $m$ if it can be expressed as the product of two integers $u\,v=m$. Ignoring memory costs, a sizable fraction of the keys can be recovered with $2^{67}$ modexps per attacked key. And while the simplest version of that attack requires unrealistically much memory, time vs memory trade-offs may be possible (that's discussed there).