In my opinion, you should pick the cell line that most closely resembles the human pathology in consideration of the phenomenon you are trying to model.
One consideration might be the Estrogen Receptor status of the cells (231's are ER-, MCF-7 are ER+), as ER status has been shown to influence metastatic site. (See references: Koenders PG, Beex LV, Langens R, Kloppenborg PW, Smals AG, Benraad TJ. Steroid hormone receptor activity of primary human breast cancer and pattern of first metastasis. The Breast Cancer Study Group. Breast Cancer Res Treat 1991; 18: 27–32. and Fuqua SA. The role of estrogen receptors in breast cancer metastasis. J Mammary Gland Biol Neoplasia 2001; 6: 407–17.) Therefore, ER+ MCF-7 cells might better recapitulate the interaction between cells in the niche environment of the bone.
On the other hand, the general metastatic potential of breast cancer cell lines tends to be higher in ER- cell lines than ER+ cell lines, particularly in vivo. If you are studying the process of metastasis (such as intravasation into the bone), ER+ cell lines may not work well.
Of course, other characteristics besides ER may also come into play in your model, but the field is frequently divided based on ER status.
Generally, MDA-MB-231 is considered the most suitable cell line model for bone metastasis. This is indeed a triple negative cell line with spontaneous metastatic activity in vivo. You can check several papers on this topic, published by the group of Joan Massagué. They even have a MDA-MB-231-BoM variant, enriched for cells with specific osteolytic bone tropism. Check Minn A et al, Nature 2005, Zhang et al, Cancer Cell 2009, etc. Or just check Massagué on pubmed, you'll find all about it.