I have a key pair, say $(d, Q)$. I send my public key $Q$ to 2 different people. I also get their public keys $Q_1$ and $Q_2$. Now I can derive 2 shared secrets, $d*Q_1$ and $d*Q_2$. The 2 other people can also derive secrets $d_1*Q$ and $d_2*Q$. Assuming all key pairs were generated from the same elliptic curve, is it safe to use both the shared secrets in the same session for 2-way communication?
Yes, you can use a (semi) static key pair for (Elliptic Curve) Diffie-Hellman. If you want to check for sure that you use ECDH correctly take a look at the NIST SP 56A which shows the various way that key agreement can be used. In this case you'd probably look for 22.214.171.124.
Note that you should check the public keys for validity. Furthermore, this describes a key agreement protocol. If the static public key can be trusted by the two other parties then the party holding the static key pair has been authenticated. You may require additional authentication for the other two parties. Finally, you probably want to use a MAC to validate that you've both generated the correct secret keys in your key agreement protocol rather than afterwards. This probably means that you need a KDF to derive the various session keys. You may want to use different keys for each direction of communication as well.